Analysis and implementation of radial basis function neural network for controlling non-linear dynamical systems
Modelling and control of non-linear systems are not easy, which are now being solved by the application of neural networks. Neural networks have been proved to solve these problems as they are described by adjustable parameters which are readily adaptable online. Many types of neural networks have been used and the most common one is the backpropagation algorithm. The algorithm has some disadvantages, such as slow convergence and construction complexity. An alternative neural networks to overcome the limitations associated with the backpropagation algorithm is the Radial Basis Function Network which has been widely used for solving many complex problems. The Radial Basis Function Network is considered in this theses, along with a new adaptive algorithm which has been developed to overcome the problem of the optimum parameter selection. Use of the new algorithm reduces the trial and error of selecting the minimum required number of centres and guarantees the optimum values of the centres, the widths between the centres and the network weights. Computer simulation usmg SimulinklMatlab packages, demonstrated the results of modelling and control of non-linear systems. Moreover, the algorithm is used for selecting the optimum parameters of a non-linear real system 'Brushless DC Motor'. In the laboratory implementation satisfactory results have been achieved, which show that the Radial Basis Function may be used for modelling and on-line control of such real non-linear systems.